JP2008171097A - Vehicle recognition apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a vehicle recognition apparatus capable of enhancing the recognizability of a subject even if there is a local high-brightness area during the night. <P>SOLUTION: The vehicle recognition apparatus 1 includes a camera 2 for capturing an image including the subject of recognition, and an image processing part 3 for ternalizing the captured image through comparison with nearby pixels using a threshold. The image processing part 3 includes a local exclusion process part 32 for dividing the captured image into a plurality of blocks, and excluding blocks of locally bright parts from the subjects of computing, and a ternalizing process part 33 for computing the best expected value of the threshold based on the brightness average and the brightness dispersion of the image from which the blocks of locally bright parts have been excluded by the local exclusion process part 32. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention belongs to the technical field of a vehicular recognition device that recognizes the number or road sign of another vehicle from an image.

2. Description of the Related Art Conventionally, a vehicle recognition device includes a camera that captures an image including a license plate of another vehicle, and a ternary processing unit that performs ternarization on the captured image by using a threshold value with a neighboring pixel. 2, the optimal prediction value of the threshold value is calculated from the luminance average and luminance variance of the captured image, and the ternary processing unit performs ternarization using the threshold prediction value (see, for example, Patent Document 1). ).
JP 2006-309650 A (page 2-7, full view)

  However, in the past, the brightness average and brightness variance are calculated from the entire image, so it is effective for the overall bright image and the overall dark image, but the headlight of the oncoming vehicle is not captured at night. There is a problem that an appropriate threshold value cannot be obtained in the case of an image having a large difference in brightness and darkness, such as an image that is blurred or an image in which a stop lamp of the front car is reflected at night.

  The present invention has been made paying attention to the above-mentioned problems, and the object of the present invention is to improve the recognizability of an object even when a local high-intensity region exists at night. An object is to provide a vehicle recognition device.

  In order to achieve the above object, in the present invention, an imaging unit that captures an image including a recognition object, an image processing unit that performs ternarization on the captured image by comparison using a threshold value with neighboring pixels, In the vehicular recognition apparatus, the image processing means includes an image dividing means for dividing the captured image into a plurality of blocks, a local high-luminance exclusion means for excluding locally bright blocks from the calculation target, Threshold value predicting means for calculating an optimum predicted value of the threshold value from the luminance average and luminance variance of the image obtained by excluding locally bright blocks by the local high luminance exclusion means.

  Therefore, in the present invention, it is possible to improve the recognizability of an object even when there is a local high luminance area at night.

  Hereinafter, an embodiment for realizing a vehicular recognition device of the present invention will be described based on a first embodiment corresponding to the first and second aspects of the invention.

First, the configuration will be described.
FIG. 1 is a schematic diagram illustrating the configuration of the vehicle recognition apparatus according to the first embodiment.
The vehicle recognition device 1 mainly includes a camera 2, an image processing unit 3, and a vehicle device 4.
The camera 2 is a CCD camera and takes a 24-bit color image and outputs it as an image data signal.
The image processing unit 3 includes a gray scale processing unit 31, a local exclusion processing unit 32, a ternary processing unit 33, and a recognition processing unit 34, and processes image data input from the camera 2.

The gray scale processing unit 31 converts the color image into a gray scale image having a luminance value of 8 bits in black and white.
The local exclusion processing unit 32 performs processing for excluding portions such as stop lamps and reflection of headlights. Details will be described later.
The ternarization processing unit 33 classifies each pixel of the remaining grayscale image that is not excluded into three types (colorless, white, and black), and generates a ternary image. In this ternary classification, a threshold f1 is used to determine white, and a threshold f2 is used to determine black.

  The thresholds f1 and f2 are obtained by using C11 to C13 and C21 to C23 as coefficients of the prediction formula.

  f1 = C11 × luminance average + C12 × luminance variance + C13 (Expression 1)

  f2 = C21 × luminance average + C22 × luminance variance + C23 (Expression 2)

What was obtained from the above formula is used.
Equations 1 and 2 are obtained in advance by experiments and multiple regression analysis.
The recognition processing unit 34 performs recognition processing by matching characters, numbers, and shapes displayed on a license plate, a sign, etc. from a ternary image, and recognizes the size of the license plate and the contents of the sign. Do.
The vehicle device 4 is a device that uses the obtained vehicle number, the size of the license plate, the contents of the sign, and the like. For example, a device that calculates the inter-vehicle distance from the size of the license plate may be used.

The operation will be described.
[Local exclusion process]
FIG. 2 is a flowchart showing a flow of image processing mainly including local exclusion processing executed by the image processing unit 3 of the vehicle recognition apparatus according to the first embodiment. Each step will be described below.

  In step S1, horizontal and vertical integrated luminances are created from the input camera image.

  In step S2, points where the slope of each integrated luminance graph is greater than or equal to a certain value are extracted.

  In step S3, a block boundary line is determined from points on the integrated luminance graph.

  In step S4, the average brightness of each block is calculated.

  In step S5, the block having the maximum luminance average is excluded.

  In step S6, blocks having a luminance average of x% or more of the maximum luminance average are excluded.

  In step S7, the screen brightness is calculated from the remaining blocks.

  In step S8, the optimal threshold value is predicted from the screen luminance by multivariate analysis, and the process proceeds to a recognition algorithm.

[Action to eliminate local high brightness area]
FIG. 3 is an explanatory diagram illustrating an image and a horizontal and vertical integrated luminance graph of the image in the vehicle recognition apparatus according to the first embodiment. FIG. 4 is an explanatory diagram illustrating a state in which the integrated luminance graphs in the horizontal and vertical directions of the image in the vehicle recognition apparatus according to the first embodiment are processed and reflected on the screen. FIG. 5 is an explanatory diagram illustrating a state in which the local high-intensity block of the image is excluded in the vehicle recognition apparatus according to the first embodiment.

In the gray scale processing unit 31 and the local exclusion processing unit 32 of the image processing unit 3 according to the first embodiment, the color image is converted into a gray scale image having a luminance value of 8 bits in black and white, and as shown in FIG. The horizontal integrated luminance graph 20 and the vertical integrated luminance graph 30 are created (step S1).
Next, as shown in FIG. 4, points at which the slopes 211 and 311 of the integrated luminance graphs 21 and 31 exceed a certain value, that is, boundary points 212 and 312 are extracted (step S2). Then, as shown in FIG. 4, block division boundary lines 213 and 313 are set from points on the integrated luminance graph (step S3).

Next, as shown in FIG. 5, with respect to the image 12 partitioned so as to be divided into a plurality of blocks by the block division boundary lines 213 and 313, the luminance average of each block is calculated (step S4).
Then, the block having the highest luminance average among all the blocks is excluded (step S5). Further, blocks having a luminance average of which the maximum luminance average is x% or more are excluded (step S6). FIG. 5 shows the excluded local high intensity block 121.

Next, the ternary processing unit 33 according to the first embodiment calculates the screen luminance from the remaining blocks other than the excluded blocks (step S7), predicts the optimum threshold from the calculated screen luminance (step S8), and then The ternarization process using the threshold value is performed.
In the first embodiment, the image is divided into blocks according to the gradient of luminance in this way, and blocks with locally high luminance are excluded. In the case of an image in which the stop lamp of the front car is reflected, the appropriate threshold value is obtained from the multiple regression equation using the same process as before by eliminating blocks that are locally high in brightness. It is possible to recognize an object to be recognized (number plate, road sign, etc.).

In the process of eliminating local high-brightness blocks before ternarization according to the first embodiment, the block boundary line is obtained from the change in brightness, and therefore the calculation cost can be kept low.
Moreover, although this process is effective at night, it can be used without any influence on daytime use, and it is not necessary to switch between daytime and nighttime.
This reduces the switching processing load, the software storage load, and the like, and leads to cost reduction.

The effect | action which excludes the local high-intensity area | region of the present Example 1 is further demonstrated.
FIG. 6 is a graph showing the relationship between the average brightness and the threshold value of the vehicle recognition apparatus according to the first embodiment and the threshold prediction formula obtained by the multiple regression equation. FIG. 7 is an explanatory diagram illustrating an example of an image processed using the threshold A in the vehicle recognition apparatus according to the first embodiment. FIG. 8 is an explanatory diagram illustrating an example of an image processed using the threshold B in the vehicle recognition apparatus according to the first embodiment.
In the original images of FIGS. 7 and 8, it is assumed that the stop lamp of the previous vehicle is lit.
Conventionally, by using a multiple regression analysis, a linear expression that predicts a threshold value for ternary processing from the luminance average and luminance variance of an image is created, thereby obtaining an effect of reducing the amount of calculation.
In this case, the graph of FIG. 6 shows the relationship between the luminance average on the horizontal axis and the threshold value on the vertical axis.

In this threshold prediction line 100, since the stop lamp of the preceding vehicle is lit, the average brightness is raised high, and the threshold B at point B is obtained.
This threshold value B is optimal when, for example, what is obtained from the image 14 in FIG. 8 is a license plate number, a stop lamp having a higher brightness than the brightness in the vicinity of the number is turned on. A threshold value B for high brightness is obtained than the threshold value. Therefore, it is difficult to read the number on the license plate in the processed image 14 of FIG. Therefore, reading is difficult even in subsequent processing such as pattern matching.

On the other hand, in the first embodiment, the local high-luminance portion due to the stop lamp lighting is deleted, and the arithmetic processing for obtaining the luminance average or the like is performed. Therefore, as shown in FIG. As compared with the threshold value A, the threshold value A of the point A for low luminance is obtained.
As a result, a threshold A for luminance closer to the number is obtained. Therefore, reference numeral 13 in the processed image in FIG. 7 can read the license plate number. Therefore, reading can be performed in subsequent processing such as pattern matching.

Next, the effect will be described.
In the vehicle recognition apparatus according to the first embodiment, the effects listed below can be obtained.
(1) In a vehicle recognition apparatus 1 including a camera 2 that captures an image including a recognition object, and an image processing unit 3 that performs ternarization on the captured image by comparison using a threshold value with neighboring pixels. The image processing unit 3 divides the captured image into a plurality of blocks, a local exclusion processing unit 32 that excludes locally bright blocks from the calculation target, and a local bright processing block by the local exclusion processing unit 32. 3 is provided with a ternary processing unit 33 that calculates an optimum predicted value of the threshold from the luminance average and luminance variance of the image excluding the image, so that even if there is a local high luminance region at night, Recognizability can be improved.

  (2) The image division process of the local exclusion processing unit 32 sets the boundary line for dividing the image from the inclination of the integrated luminance graph in the horizontal direction of the image and the inclination of the integrated luminance graph in the vertical direction of the image. Since the block boundary line is obtained from the change, the calculation cost can be kept low.

  As mentioned above, although the vehicle recognition apparatus of the present invention has been described based on the first embodiment, the specific configuration is not limited to these embodiments, and the invention according to each claim of the claims. Design changes and additions are allowed without departing from the gist.

It is the schematic which shows the structure of the recognition apparatus for vehicles of Example 1. FIG. 4 is a flowchart illustrating a flow of image processing mainly including local exclusion processing executed by the image processing unit 3 of the vehicle recognition apparatus according to the first embodiment. It is explanatory drawing which shows the integrated luminance graph of the image in the vehicle recognition apparatus of Example 1, and the horizontal and vertical direction of the image. It is explanatory drawing which shows the state which processed the horizontal and vertical integrated luminance graph of the image in the vehicle recognition apparatus of Example 1, and reflected it on the screen. It is explanatory drawing which shows the state which excluded the local high-intensity block of the image in the vehicle recognition apparatus of Example 1. FIG. It is a graph which shows the threshold value prediction formula calculated | required by the relationship between the brightness | luminance average of the vehicle recognition apparatus of Example 1, and a threshold value, and multiple regression. FIG. 6 is an explanatory diagram illustrating an example of an image processed using a threshold value A in the vehicle recognition apparatus according to the first embodiment. FIG. 5 is an explanatory diagram illustrating an example of an image processed using a threshold value B in the vehicle recognition apparatus according to the first embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Vehicle recognition apparatus 2 Camera 3 Image processing part 31 Gray scale process part 32 Local exclusion process part 33 Trinarization process part 34 Recognition process part 4 Vehicle apparatus 10 Image 11 Image 12 Image 13 Image 14 Image 20 Integrated luminance graph 21 Integration Luminance graph 211 Inclination 212 Boundary point 213 Block division boundary line 30 Integrated luminance graph 31 Integrated luminance graph 311 Inclination 312 Boundary point 313 Block division boundary line 100 Threshold prediction line 121 Local high-intensity block A Threshold B Threshold

Claims (2)

Imaging means for capturing an image including a recognition object;
Image processing means for performing ternarization on a captured image by comparison using a threshold value with neighboring pixels;
In a vehicle recognition device comprising:
The image processing means includes
Image dividing means for dividing the captured image into a plurality of blocks;
A local high-intensity exclusion means for excluding locally bright blocks from the calculation target;
Threshold prediction means for calculating an optimum predicted value of the threshold from the luminance average and luminance variance of the image excluding the locally bright block by the local high luminance exclusion means;
A vehicle recognition device comprising: The vehicle recognition device according to claim 1,
The image dividing means includes
Set the boundary line that divides the image from the inclination of the integrated luminance graph in the horizontal direction of the image and the inclination of the integrated luminance graph in the vertical direction of the image,
A recognition apparatus for a vehicle.